Databricks launches advanced language model DBRX, surpassing open-source models in various benchmarks. DBRX outperforms GPT-3.5, showing potential for open-source alternatives. Its MoE architecture balances performance and speed with 132B parameters. Users can access DBRX via Databricks platform, GitHub, and Hugging Face for AI application enhancement.

A recent study in Neurology found that GPT-4 accurately localized acute stroke lesions based on clinical data. The model showed high precision and detailed reasoning, indicating potential for assisting healthcare providers in neuroanatomical localization.

Rabbit unveils r1 AI device with lifelike voice interaction capabilities in collaboration with ElevenLabs. Features include LAM, real-time search, bidirectional translation, and AI summaries. Updates and new features expected in the future. Other players like Humane entering the AI market.

Large Language Models (LLMs) have great potential in finance, improving efficiency, security, and customer service. Professionals are adopting LLMs for tasks like regulatory review and anticipate integration into services like investment banking. Collaboration is recommended to address security concerns.

Google introduced Gemini Nano in December as a mobile-optimized language model for Android. Pixel 8 Pro is the only device currently supporting it due to hardware limitations, raising questions among users and experts. Samsung, on the other hand, has started rolling out AI features on its previous flagship phones. This discrepancy highlights the importance of consumer advocacy in driving technology advancements.

SenseTime achieved milestones in large-scale devices and models, with over 12,000P operating power and 45,000 GPUs. Generative AI business revenue soared, showing a shift away from traditional AI business. Future goals include maintaining tech leadership, expanding computing power, and driving generative AI commercialization.

AI21 Labs releases Jamba, a production-grade AI model based on Mamba architecture, combining Mamba SSM and Transformer. Boasting 256K token context window, Jamba outperforms Transformer-based models like Mixtral 8x7B, promising high throughput and efficiency. Released under Apache 2.0, Jamba shows impressive results, tailored for enterprise deployment via NVIDIA NIM Inference Microservice. Stay tuned for a more secure commercial version.

OpenAI emphasizes text-based models as the foundation for intelligent AI and possibly AGI. Current focus is on developing an LLM OS integrating various modalities. Debates continue on the role of language in AI and AGI development.

Osmo, an AI startup, is developing "smell transmission" tech to capture, analyze, and replicate smells for long-distance transmission. Founded by former Google "Brain" researcher, Alex Wiltschko, the company aims to bring smell into the digital age.

Adobe Firefly introduces new "Structure Reference" feature for applying existing image layouts to generated images, enhancing creative control. Users can easily create multiple variations with the same layout and style. This innovative tool also promotes transparency and safe commercial use, training on licensed and public domain content. Structure Reference now available to all Firefly subscribers.

Elon Musk's xAI opens revolutionary Grok chatbot to more Premium subscribers, aiming to compete with industry giants like OpenAI and Anthropic. Facing declining user numbers, X platform provides access to AI chatbots to retain users and increase revenue. Grok's ability to answer diverse topics sets it apart, but X must address user engagement to ensure its success.

Large language models (LLMs) are evolving into multimodal large language models (MLLMs) to handle various forms of information. HyperLLaVA, a dynamic version of LLaVA, utilizes expert modules to generate dynamic parameters for improved reasoning in multimodal tasks. The innovative approach outperforms existing methods in benchmark tests, showcasing its potential for advancing multimodal learning systems.

On March 27th, Baidu Intelligent Cloud launched 7 innovative products covering various scenarios like marketing, customer service, and knowledge management. These SaaS products offer enterprises flexible solutions for digital upgrades and efficient operations.

NVIDIA dominates in generative AI with 3x performance boost on GPT-J LLM using TensorRT-LLM. Hopper GPU achieves record throughput in MLPerf test, paving the way for innovative AI advancements. Future Blackwell architecture GPU aims to support trillion-parameter AI models.

MLPerf Inference v4.0 introduces larger generative AI models like Llama 2 70B and Stable Diffusion XL, showcasing performance improvements. NVIDIA's Hopper system with TensorRT-LLM software leads in generative AI, while Intel offers cost-effective alternatives with Gaudi 2 and 5th gen Xeon processors. Industry moves towards trillion-parameter AI models as hardware and software vendors compete for performance.

Amazon completes $4 billion investment in AI startup Anthropic, securing minority stake. Anthropic, founded by ex-OpenAI researchers, leads in generative AI with Claude chatbots. Partnership with AWS boosts cloud services for future AI models, challenging OpenAI and Google.

OpenAI collaborates with US developers to test GPT earnings. GPT Store for users launched, allowing customization and sharing of AI agents. Developers to be compensated based on user interactions. Over 3 million customized ChatGPT versions generated. Find GPTs in various categories at chat.openai.com/gpts. ChatGPT Team launched for team collaboration.

MediaTek deployed T-Brain 1000 model on Dimensity 9300, enabling integration of large-scale models on mobile chips. Collaboration with Alibaba Cloud to provide edge-side solutions for smartphone manufacturers. T-Brain 1000 open-sourced various large language models. MediaTek also developing own large language models. Collaboration aims to enhance smartphone intelligence and user experience. The future holds more advanced intelligent solutions for smartphone industry.

Samsung plans to launch Mach-1 to compete in AI market against NVIDIA. With AGI lab support, Samsung aims to improve power efficiency by utilizing new algorithms to reduce GPU and memory bottleneck. Mach-1 prototype expected by 2024 end.

OpenAI is developing a new speech engine project, focusing on speech recognition and natural interactions with devices. This innovative project aims to revolutionize AI assistants, potentially leading to the most advanced personal assistants in the tech industry. By seamlessly integrating with smart devices, it could change our perception and usage of AI, setting new benchmarks for personal assistants. This project is anticipated to have a significant impact on the future of AI.